No abstract

No abstract

There are several applications that require automatic text classification or categorization. Several approaches and algorithms are proposed to predict a document category based on predefined ones. Examples of such algorithms include: N-gram, Manhattan, Dice and Euclidean similarity measures. This paper includes an extensive evaluation for N-gram possible N-options (i.e. 3, 4 and 5 characters selection). Comparison is also made based on evaluating the effect of preprocessing and dataset training on the quality and accuracy of prediction. Results showed that for Arabic text categorization, it is best to use 3-letters N-gram then 5-letters N-gram and finally 4-letters N-gram. Preprocessing of stop words removal did not show a significant improvement on the precision of classes' prediction. Further, using a larger training dataset showed a significant improvement of prediction accuracy. In terms of similarity measures, Euclidean is shown to be the best of those evaluated for document classification then Dice and finally Manhattan.

No abstract